Upload metrics.py with huggingface_hub
Browse files- metrics.py +41 -2
metrics.py
CHANGED
@@ -8,6 +8,7 @@ import evaluate
|
|
8 |
import nltk
|
9 |
import numpy
|
10 |
|
|
|
11 |
from .operator import (
|
12 |
MultiStreamOperator,
|
13 |
SingleStreamOperator,
|
@@ -60,7 +61,13 @@ class GlobalMetric(SingleStreamOperator, Metric):
|
|
60 |
|
61 |
refs, pred = instance["references"], instance["prediction"]
|
62 |
|
63 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
64 |
instance["score"]["instance"].update(instance_score)
|
65 |
|
66 |
references.append(refs)
|
@@ -355,8 +362,27 @@ class Bleu(HuggingfaceMetric):
|
|
355 |
scale = 1.0
|
356 |
|
357 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
358 |
class CustomF1(GlobalMetric):
|
359 |
main_score = "f1_micro"
|
|
|
360 |
|
361 |
@abstractmethod
|
362 |
def get_element_group(self, element):
|
@@ -391,6 +417,10 @@ class CustomF1(GlobalMetric):
|
|
391 |
assert len(references) == len(predictions), (
|
392 |
f"references size ({len(references)})" f" doesn't mach predictions sise ({len(references)})."
|
393 |
)
|
|
|
|
|
|
|
|
|
394 |
groups_statistics = dict()
|
395 |
for references_batch, predictions_batch in zip(references, predictions):
|
396 |
grouped_references = self.group_elements(references_batch)
|
@@ -418,6 +448,7 @@ class CustomF1(GlobalMetric):
|
|
418 |
groups_statistics[group]["recall_denominator"] += rd
|
419 |
|
420 |
result = {}
|
|
|
421 |
pn_total = pd_total = rn_total = rd_total = 0
|
422 |
for group in groups_statistics.keys():
|
423 |
pn, pd, rn, rd = (
|
@@ -426,13 +457,21 @@ class CustomF1(GlobalMetric):
|
|
426 |
groups_statistics[group]["recall_numerator"],
|
427 |
groups_statistics[group]["recall_denominator"],
|
428 |
)
|
429 |
-
result[f"f1_{group}"] = self.f1(pn, pd, rn, rd)
|
430 |
pn_total, pd_total, rn_total, rd_total = pn_total + pn, pd_total + pd, rn_total + rn, rd_total + rd
|
|
|
|
|
|
|
|
|
431 |
try:
|
432 |
result["f1_macro"] = sum(result.values()) / len(result.keys())
|
433 |
except ZeroDivisionError:
|
434 |
result["f1_macro"] = 1.0
|
435 |
|
|
|
|
|
|
|
|
|
|
|
436 |
result[f"f1_micro"] = self.f1(pn_total, pd_total, rn_total, rd_total)
|
437 |
return result
|
438 |
|
|
|
8 |
import nltk
|
9 |
import numpy
|
10 |
|
11 |
+
from .dataclass import InternalField
|
12 |
from .operator import (
|
13 |
MultiStreamOperator,
|
14 |
SingleStreamOperator,
|
|
|
61 |
|
62 |
refs, pred = instance["references"], instance["prediction"]
|
63 |
|
64 |
+
try:
|
65 |
+
instance_score = self._compute([refs], [pred])
|
66 |
+
except:
|
67 |
+
instance_score = {"score": None}
|
68 |
+
if isinstance(self.main_score, str) and self.main_score is not None:
|
69 |
+
instance_score[self.main_score] = None
|
70 |
+
|
71 |
instance["score"]["instance"].update(instance_score)
|
72 |
|
73 |
references.append(refs)
|
|
|
362 |
scale = 1.0
|
363 |
|
364 |
|
365 |
+
class MatthewsCorrelation(HuggingfaceMetric):
|
366 |
+
metric_name = "matthews_correlation"
|
367 |
+
main_score = "matthews_correlation"
|
368 |
+
str_to_id: dict = InternalField(default_factory=dict)
|
369 |
+
|
370 |
+
def get_str_id(self, str):
|
371 |
+
if str not in self.str_to_id:
|
372 |
+
id = len(self.str_to_id)
|
373 |
+
self.str_to_id[str] = id
|
374 |
+
return self.str_to_id[str]
|
375 |
+
|
376 |
+
def compute(self, references: List[List[str]], predictions: List[str]) -> dict:
|
377 |
+
formatted_references = [self.get_str_id(reference[0]) for reference in references]
|
378 |
+
formatted_predictions = [self.get_str_id(prediction) for prediction in predictions]
|
379 |
+
result = self.metric.compute(predictions=formatted_predictions, references=formatted_references)
|
380 |
+
return result
|
381 |
+
|
382 |
+
|
383 |
class CustomF1(GlobalMetric):
|
384 |
main_score = "f1_micro"
|
385 |
+
classes = None
|
386 |
|
387 |
@abstractmethod
|
388 |
def get_element_group(self, element):
|
|
|
417 |
assert len(references) == len(predictions), (
|
418 |
f"references size ({len(references)})" f" doesn't mach predictions sise ({len(references)})."
|
419 |
)
|
420 |
+
if self.classes is None:
|
421 |
+
classes = set([self.get_element_group(e) for sublist in references for e in sublist])
|
422 |
+
else:
|
423 |
+
classes = self.classes
|
424 |
groups_statistics = dict()
|
425 |
for references_batch, predictions_batch in zip(references, predictions):
|
426 |
grouped_references = self.group_elements(references_batch)
|
|
|
448 |
groups_statistics[group]["recall_denominator"] += rd
|
449 |
|
450 |
result = {}
|
451 |
+
num_of_unknown_class_predictions = 0
|
452 |
pn_total = pd_total = rn_total = rd_total = 0
|
453 |
for group in groups_statistics.keys():
|
454 |
pn, pd, rn, rd = (
|
|
|
457 |
groups_statistics[group]["recall_numerator"],
|
458 |
groups_statistics[group]["recall_denominator"],
|
459 |
)
|
|
|
460 |
pn_total, pd_total, rn_total, rd_total = pn_total + pn, pd_total + pd, rn_total + rn, rd_total + rd
|
461 |
+
if group in classes:
|
462 |
+
result[f"f1_{group}"] = self.f1(pn, pd, rn, rd)
|
463 |
+
else:
|
464 |
+
num_of_unknown_class_predictions += pd
|
465 |
try:
|
466 |
result["f1_macro"] = sum(result.values()) / len(result.keys())
|
467 |
except ZeroDivisionError:
|
468 |
result["f1_macro"] = 1.0
|
469 |
|
470 |
+
amount_of_predictions = pd_total
|
471 |
+
if amount_of_predictions == 0:
|
472 |
+
result["in_classes_support"] = 1.0
|
473 |
+
else:
|
474 |
+
result["in_classes_support"] = 1.0 - num_of_unknown_class_predictions / amount_of_predictions
|
475 |
result[f"f1_micro"] = self.f1(pn_total, pd_total, rn_total, rd_total)
|
476 |
return result
|
477 |
|